[clang] [clang][Sema] Add checks for validity of default ctor's class (PR #78898)

Vlad Serebrennikov via cfe-commits cfe-commits at lists.llvm.org
Mon Jan 22 06:39:30 PST 2024


Endilll wrote:

> why can't we add -verify test? Yes, it will be checking errors that the patch didn't touch, but it is what mostly people do when adding clang tests and it will be +N test cases in a regular test base which not only ensure your change is correct, but the future ones too.

Because we [have](https://github.com/llvm/llvm-project/issues?q=is%3Aissue+is%3Aclosed+reason%3Acompleted+label%3Acrash%2Ccrash-on-valid%2Ccrash-on-invalid+label%3Aclang%3Afrontend%2Cclang%3Acodegen%2Cclang%3Adiagnostics%2Cclang%3APCH%2Cclang%3Aheaders+-label%3Aduplicate%2Cinvalid%2Cincomplete%2Cwontfix%2Cworksforme) more than a couple of tests that should ensure we don't crash. Many of them don't make much sense, e.g. automatically reduced ones (even if identifiers are kept intact; #67914 is not a bad example). Which makes it harder for people altering diagnostic behavior to reason about such tests when they fail to pass CI. It also makes contributors to have more "fun" like @AaronBallman [had](https://github.com/llvm/llvm-project/commit/84a3aadf0f2483dde0acfc4e79f2a075a5f35bd1) no so long ago, updating tons of tests because of his VLA-related change.

Overspecifying a significant corpus of hard-to-reason-about tests leads us to a slippery slope of a applying changes blindly and developing scripts akin to `update_cc_test_checks.py`. I see a significant amount of harm in this direction.

https://github.com/llvm/llvm-project/pull/78898


More information about the cfe-commits mailing list